Evaluating Routine Variability of Daily Activities in Smart Homes with Image Complexity Measures

Abstract

With the increasing trend of older adults living alone, an efficient and nonintrusive way to monitor these individuals’ mental health status is required for early diagnosis of mental disease (e.g., dementia). Because the routine variability of activities of daily living (ADLs) can act as an index of an older adult’s mental status, various research has attempted to develop a metric that can quantify and measure ADL routine variability. However, this research has focused either on the assessment of a single key ADL or the differences between activities performed on consecutive days. These approaches cannot measure the periodic changes over the long term (e.g., when the performed routine is different for each day of the week, or when exceptional events occurred) that may reflect mental health status. This study hypothesizes that the level of image complexity of the visualized data in ADL logs can represent the level of routine variability. To test this hypothesis, synthetic images are designed and generated presenting various randomness in ADL routines [i.e., varying the duration of routine-contributing activities (RADLs), varying the number of occurrences of non-routine-contributing activities (NRADLs), and varying the order of RADLs]. The correlations between each type of variability and the complexity values are identified, and values from canny edge distance and contrast of gray-level co-occurrence matrix (GLCM) elements show a strong correlation with the variability in the visualized data of the ADL logs. The metrics are further evaluated by using data collected from older adults living alone in real residential environments. This study suggests that image complexity metrics can be used to track gradual changes in routine variability within a subject, laying the foundation toward unobtrusive longitudinal measures of cognitive decline that could lead to the early prognosis of mental diseases.